An adaptive spectral line enhancement method for acoustic emission signals based on Hurst exponent

Abstract

This invention discloses an adaptive spectral line enhancement method for acoustic emission signals based on the Hurst exponent, belonging to the field of nondestructive testing. The Hurst exponent is introduced into the adaptive spectral line enhancer, and the adaptive weights are iteratively updated by minimizing the Hurst exponent. A multi-parameter fuzzy evaluation model and a genetic algorithm are used to select the optimal parameter combination of the filter. The parameter-optimized adaptive spectral line enhancer is then applied to the processing of typical strong-noise acoustic emission signals. This invention's method, by introducing the Hurst exponent into the adaptive spectral line enhancer, can accelerate the convergence speed of the filter. Simultaneously, the use of multi-parameter fuzzy evaluation and a genetic algorithm to optimize the filter parameter combination improves its filtering effect.

Description

Technical Field

[0001] This invention relates to an improved adaptive line spectrum enhancement method based on the Hurst exponent, which is suitable for extracting acoustic emission signals under strong noise conditions and belongs to the field of non-destructive testing. Background Technology

[0002] Metal plates can deform or break under external forces, often accompanied by the generation of acoustic emission signals. Various signal analysis methods have been applied to try to extract acoustic emission signals, but when the acoustic emission signals are mixed with a large amount of noise, signal extraction is greatly affected.

[0003] Domestic and international scholars have attempted to apply various signal analysis methods to acoustic emission signal extraction, such as wavelet transform and empirical mode decomposition. Traditional non-stationary signal processing methods can effectively suppress unstructured noise, but their suppression of structured noise (structural vibration, factory noise, etc.) is not ideal. While sparse representation can extract signals in complex environments through sparse decomposition and sparse reconstruction, it often performs poorly in situations with strong system noise. Therefore, signal extraction methods based on specific features have emerged as a promising approach. (Dan et al.) [1] A signal extraction method based on Tsallis synchronous extrusion wavelet entropy is proposed. This method uses Tsallis entropy to more accurately quantify the local changes of acoustic emission wavelet coefficients, improving the signal-to-noise ratio and making it easier to effectively detect and locate cracks. (Xin et al.) [2] Discrete wavelet transform is applied to simulated contaminated rail crack signals. Then, the acoustic emission signals of rail cracks are extracted by utilizing the relationship between wavelet coefficients and Shannon entropy, thus realizing the localization of rail crack acoustic emission signals.

[0004] To address the extraction of acoustic emission signals in noisy environments, this invention proposes an adaptive spectral enhancement method for acoustic emission signals based on the Hurst exponent. This invention incorporates the Hurst exponent into the cost function of the adaptive filter, utilizing the minimization of the Hurst exponent and Lagrange multipliers to iteratively update the adaptive weights, thereby achieving rapid adaptive convergence and signal enhancement. Summary of the Invention

[0005] The present invention proposes an adaptive spectral line enhancement method for acoustic emission signals based on the Hurst exponent, the structure of which is as follows: Figure 1 As shown, its basic principle is:

[0006] 1) Conduct fractal characteristic analysis of acoustic emission noise signals. Calculate the mean and standard deviation of the noise signal at different scale increments to determine whether the noise signal conforms to fractal Brownian motion, i.e., the increment Y of the noise signal... m A Gaussian signal with zero mean, and Ym The Gaussian statistical characteristics are basically consistent with the normal distribution.

[0007] If the noise signal conforms to Brownian motion, calculate its Hurst exponent.

[0008]

[0009] Where α1 is the time shift of the acoustic emission signal 1, and α2 is the time shift of the acoustic emission signal 2.

[0010] 2) An adaptive spectral line enhancement method based on the Hurst exponent is proposed, such as... Figure 1 As shown, the Hurst exponent and Lagrange multipliers are introduced into the cost function of the adaptive filter. By delaying the input data samples of the adaptive line spectrum enhancer, the broadband noise components in the data samples before and after the delay are uncorrelated, while the line spectrum components remain correlated. The mean square error, Hurst exponent, and Lagrange multipliers are minimized to perform adaptive iteration of the weights, achieving fast adaptive convergence and signal enhancement.

[0011] Constructed cost function Q w for:

[0012] Q w =E{e 2 (n)}+λH (2)

[0013] Where λ is the Lagrange multiplier and H is the Hurst exponent.

[0014] Using stochastic gradient descent to analyze Q w Based on the solution, we obtain:

[0015]

[0016] Where i1 is the acoustic emission signal offset 1 and i2 is the acoustic emission signal offset 2.

[0017] The final recursive equation for the cost function is:

[0018]

[0019] Where μ is the iteration step size, and i1 and i2 are the same as in equation (3).

[0020] 3) The relative error of the focusing distance of the time-reversed acoustic emission signal and the Hurst exponent are used as characteristic parameters to evaluate the noise reduction effect of the adaptive spectral enhancement technology.

[0021] The membership function μ of the slightly smaller trapezoidal distribution of the relative error of the focusing distance of the time-reversed acoustic emission signal T δ (xδ,j ) is represented as:

[0022]

[0023] Where, x δ,j Let be the relative error index value of the convergence time of the j-th orthogonal experiment (1≤j≤25).

[0024] The membership function μ of the Hurst exponent H of a skewed normal distribution T δ (x δ,j ) is represented as:

[0025]

[0026] Where, x H,j Let be the Hurst exponent value of the j-th orthogonal experiment (1≤j≤25).

[0027] 4) Establishment of a comprehensive fuzzy evaluation model combining relative error of clustering distance and Hurst exponent.

[0028] To comprehensively consider the influence of filter order L, Lagrange multiplier λ, and two time offsets on the acoustic emission signal extraction performance, an orthogonal experimental design was first employed, followed by the use of fuzzy theory to weight and quantize the two indicators, establishing a comprehensive fuzzy signal extraction evaluation parameter model containing four feature indicators. Figure 2 A flowchart of the method is provided.

[0029] Based on the membership function values ​​of the evaluation indicators from the orthogonal experiment, the average membership function values ​​of each evaluation indicator are calculated and weighted accordingly to obtain a comprehensive fuzzy evaluation model containing two evaluation indicators:

[0030] f(x) = ω1·μ T δ (x δ,j )+ω2·μ N H (x H,j (7)

[0031] Where, μ T δ (x δ,j ) represents the membership function of a slightly smaller trapezoidal distribution for the relative error δ of the aggregation distance, μ N δ (x δ,j ) is the membership function of the Hurst exponent H as a skewed normal distribution.

[0032] 5) Through regression analysis, the influence of the four filter parameters on the filtering effect is obtained, and the optimal parameter combination of the adaptive spectral enhancer is optimized using a genetic algorithm. In the genetic algorithm optimization, the fitness function is given by equation (7).

[0033] The present invention has the following advantages: (1) Introducing the Hurst exponent into the adaptive spectral line enhancer can accelerate the convergence speed of the filter. (2) Utilizing multi-parameter fuzzy evaluation and genetic algorithm to optimize the filter parameter combination improves the filtering effect of the acoustic emission signal adaptive spectral line enhancer based on the Hurst exponent. Attached Figure Description

[0034] Figure 1 This is a diagram of the structure of an adaptive spectral line enhancer based on the Hurst exponent.

[0035] Figure 2 A process for establishing a parameter model to evaluate the comprehensive fuzzy signal extraction effect.

[0036] Figure 3 Composition of a fiber optic grating acoustic emission detection system.

[0037] Figure 4 This is the reflection spectrum of a fiber Bragg grating sensor.

[0038] Figure 5 This is a schematic diagram of a large incident angle acoustic emission localization experiment.

[0039] Figure 6 The lead at point A1 is broken, and signals are received by S1 and S2.

[0040] Figure 7 This is the time-reversed result of the A1-S1 signal.

[0041] Figure 8 This is the binarized result of the amplitude ratio difference.

[0042] Figure 9 The image shows the location results of acoustic emission at the incident angle and a magnified view of a part of it. Detailed Implementation

[0043] The invention will be further illustrated below with specific experiments:

[0044] Combination Figures 1-9 Using a fiber optic grating acoustic emission detection system, taking the location of the acoustic emission source of a broken lead signal at a large incident angle on a board as an example, this paper details the application of the improved adaptive line spectrum enhancement method based on the Hurst exponent to the extraction of acoustic emission signals under strong noise conditions.

[0045] The specific implementation steps are given below.

[0046] 1. According to Figure 3The system block diagram shown illustrates the construction of a fiber Bragg grating acoustic emission detection system. The fiber Bragg grating is attached to an aluminum plate using a remote bonding method, where the optical fibers near the grating area are bonded to the aluminum plate. The instruments and components used in the detection system are listed in Table 1.

[0047] Table 1. Instruments and components used in the detection system

[0048]

[0049] The reflection spectrum of the fiber Bragg grating sensor used is as follows: Figure 1 As shown, the center wavelengths of the two fiber Bragg grating sensors are 1539.36nm and 1539.67nm, respectively, and the -3dB bandwidth is 0.7nm for both.

[0050] 2. To demonstrate that the improved adaptive line spectrum enhancement method based on the Hurst exponent is suitable for extracting acoustic emission signals under strong noise conditions, a large incident angle lead breaking experiment was conducted on the board. The experimental schematic is shown below. Figure 5 As shown, the received signal diagram is as follows: Figure 6 As shown in Table 2.

[0051] Table 2. Location information for acoustic emission localization experiments at large incident angles.

[0052]

[0053] 3. The effects of four characteristic parameters of the filter (filter order, Lagrange multipliers, and two time offsets) on the filtering effect of the adaptive spectral line enhancer were analyzed using the orthogonal experimental method.

[0054] Using the factor levels in Table 3, an orthogonal experiment was designed. Table 4 shows the specific values ​​of the four influencing factors for different experimental combinations.

[0055] Table 3. Different factor levels in orthogonal experiments

[0056]

[0057]

[0058] Table 4. Different combinations of factors in orthogonal experiments

[0059]

[0060] Based on fuzzy theory, a fuzzy evaluation model was established, incorporating two characteristic indices: the relative error of the convergence distance of time-reversed acoustic emission signals and the Hurst exponent. The comprehensive fuzzy evaluation model, including both indices, is as follows:

[0061]

[0062] The comprehensive evaluation index values ​​for each group of orthogonal experiments are shown in Table 5.

[0063] Table 5. Comprehensive evaluation index values ​​for each group of orthogonal experiments

[0064]

[0065] 4. To achieve the best signal extraction effect, a genetic algorithm was used to optimize the four parameters proposed earlier. Regression analysis was conducted to obtain the influence of the four filter parameters on the filtering effect, and the optimal parameter combination for the adaptive spectral enhancer was optimized using a genetic algorithm. The feasible region of the parameters to be optimized is shown in Table 6, the relevant parameter settings of the genetic algorithm are shown in Table 7, and the globally optimal parameter combination obtained after genetic algorithm parameter optimization is shown in Table 8.

[0066] Table 6. Optimization Parameters for Genetic Algorithm

[0067]

[0068] Table 7 Genetic Algorithm Parameter Configuration

[0069]

[0070]

[0071] Table 8 Optimal Parameter Combinations After Optimization

[0072]

[0073] 5. Using the proposed adaptive spectral line enhancer based on the Hurst exponent and the optimal parameter combination, typical strong-noise acoustic emission signals are processed and extracted. The extracted signals are then applied to a time-reversal localization method for acoustic emission signals based on fiber grating directional characteristics. The time-reversal result of the A1-S1 signal is shown below. Figure 7 As shown. Figure 7 As shown in a), the time-reversed wave field is obtained by extracting the peak value of the reversed signal corresponding to each reversal distance in the time-reversed wave field. Figure 7 The result of b) Figure 7 In b), the x-coordinate corresponding to the highest point of the curve is the time reversal convergence distance, and the reversal result corresponding to this distance is as follows: Figure 7 As shown in c), Figure 7 The horizontal axis corresponding to the peak value in c) is the time reversal convergence moment. The amplitude spatial distribution at the time reversal convergence moment is extracted as follows: Figure 7 As shown in d).

[0074] 6. At the moment of signal convergence, the spatial distribution of amplitude is superimposed within the positioning area. The difference in directional characteristic amplitude ratio is calculated and binarized to obtain the following result: Figure 8 The results are shown.

[0075] 7. Figure 8 The result in the first step is multiplied by the acoustic emission localization result based on direction characteristics to obtain the following: Figure 9 The image result shown. (By...) Figure 9 It can be seen that the acoustic emission signal extracted using the improved adaptive line spectrum enhancement method based on Hurst exponent not only has the bright area after amplitude superposition appearing at the accurate position, but also the directional characteristics of the vertical position make the bright area in the image result more concentrated.

[0076] 8. Take the point with the highest amplitude in the image as the positioning result. The positioning results of the three acoustic emission sources are shown in Table 9. The distance ΔL between the positioning coordinates and the actual coordinates is taken as the positioning error.

[0077] Table 9. Statistical Table of Acoustic Emission Location Results at Large Incident Angles

[0078]

[0079] The positioning results show that the acoustic emission signal with a large incident angle processed by the improved adaptive line spectrum enhancement method based on the Hurst exponent can not only complete the acoustic emission positioning, but also the distance error of the positioning results is within 15mm. This proves that the improved adaptive line spectrum enhancement method based on the Hurst exponent is suitable for the extraction of acoustic emission signals under strong noise conditions.

[0080] The above is a typical application of the present invention, but the applications of the present invention are not limited thereto.

[0081] References

[0082] [1]Li D,Kuang K,Chan G K.Rail crack monitoring based on Tsallissynchrosqueezed wavelet entropy of acoustic emission signals:A field study[J].

[0083] Structural health monitoring,2018,17(6):1410-1424.

[0084] [2]Xin Z,Cui Y,Yan W,et al.An improved AE detection method of raildefect based on multi-level ANC with VSS-LMS[J].Mechanical Systems&SignalProcessing,2018,99(15):420-433.

Claims

1. A method for adaptive spectral enhancement of acoustic emission signals based on the Hurst exponent, characterized in that, Includes the following steps: 1) Analyze the noise fractal characteristics of the acoustic emission signal; Calculate the mean and standard deviation of the noise signal at different scale increments to determine if it conforms to fractal Brownian motion, i.e., the increment Y of the noise signal. m A Gaussian signal with zero mean, and Y m The Gaussian statistical characteristics are basically consistent with the normal distribution; If the noise signal conforms to Brownian motion, calculate its Hurst exponent. Where α1 is the time shift of the acoustic emission signal 1, and α2 is the time shift of the acoustic emission signal 2; 2) By incorporating the Hurst exponent into the cost function of the adaptive filter, an adaptive spectral enhancement method based on the Hurst exponent is proposed; the constructed cost function Q... w for: Q w =E{e 2 (n)}+λH (2) Where λ is the Lagrange multiplier and H is the Hurst exponent of the noise signal of the acoustic emission signal at a large incident angle; The cost function is solved using stochastic gradient descent. Where λ is the Lagrange multiplier, i1 is the acoustic emission signal offset 1, and i2 is the acoustic emission signal offset 2; The final recursive equation for the cost function is: Where μ is the iteration step size, i1 is the acoustic emission signal offset 1, and i2 is the acoustic emission signal offset 2; The main body of the adaptive line enhancer based on the Hurst exponent includes a delay K and an adaptive finite impulse response (FIR) filter. By delaying the input data samples of the adaptive line enhancer, the broadband noise components in the data samples before and after the delay are uncorrelated, while the line spectrum components remain correlated. The adaptive iteration of the weights is performed by minimizing the mean square error, the Hurst exponent, and the Lagrange multiplier, so as to achieve fast adaptive convergence and signal enhancement. 3) Analyze the effects of four characteristic parameters of the filter, namely the filter order, Lagrange multipliers, and two time offsets, on the filtering effect of the adaptive spectral line enhancer using the orthogonal experimental method; based on fuzzy theory, establish a fuzzy evaluation model that includes two characteristic indices: the relative error of the time-reversed acoustic emission signal convergence distance and the Hurst exponent. 4) Through regression analysis, the influence of the four filter parameters on the filtering effect is obtained, and the optimal parameter combination of the adaptive spectral enhancer is optimized using a genetic algorithm. 5) Using the proposed adaptive spectral enhancer based on the Hurst exponent and the optimal parameter combination, typical strong noise acoustic emission signals are processed, and the filtering effect is evaluated.

2. The method for adaptive spectral enhancement of acoustic emission signals based on Hurst exponent as described in claim 1, characterized in that: The Hurst exponent is introduced into the adaptive spectral enhancer. By minimizing the Hurst exponent, the adaptive weights are iteratively updated to achieve rapid adaptive convergence and signal enhancement.

3. The method for adaptive spectral enhancement of acoustic emission signals based on Hurst exponent as described in claim 1, characterized in that: The optimal parameter combination of the filter is selected by using a multi-parameter fuzzy evaluation model and a genetic algorithm. The parameter-optimized adaptive spectral line enhancer is then applied to the processing of typical strong noise acoustic emission signals.

4. The method for adaptive spectral enhancement of acoustic emission signals based on Hurst exponent as described in claim 1, characterized in that: Introducing the Hurst exponent into the adaptive spectral line enhancer accelerates the convergence speed of the filter. At the same time, multi-parameter fuzzy evaluation and genetic algorithm are used to optimize the filter parameter combination and improve its filtering effect.

Images